Communication networks (including Internet) enable the exchange of information and other information resources between different individuals and organizations. Generally, a network concerns the technologies of path, transmission, signaling and network management, etc. Such technologies have been widely set forth in various documents among which Telecommunications Convergence (McGraw-Hill, 2000) by Steven Shepherd, The Essential Guide to Telecommunications, Third Edition (Prentice Hall PRT, 2001) by Annabel Z. Dodd, or Communications Systems and Networks, Second Edition (M&T Books, 2000) by Ray Horak gives an overview of the technologies. The progression in such technologies obtained in the past has fully built up the speed and quality of information transmission and lowered the cost thereof.
The path technology for connecting a terminal to a wide area transmission network (for example, a local area loop of a terminal apparatus and the network edge) has been developed from a modem of 14.4, 28.8 and 56K to technologies including ISDN, T1, cable modem, DSL, Ethernet and wireless connection.
At present, transmission technologies used in a wide area network include: synchronous optical network (SONET), Dense Wavelength Division Multiplexing (DWDM), Frame Relay, Asynchronous Transmission Mode (ATM) and Resilient Packet Ring (RPR).
Among all the different signaling technologies (for example, protocols and methods for establishing, maintaining and terminating a communication in a network), Internet Protocol (IP) is applied most widely. In fact, almost all the communication and network specialists consider that an IP-based network (for example, Internet) that integrates audio (for example, telephone), video and data networks is an inevitable trend. Just as described by an author: there's one thing that is clear, that is, an IP-based train that integrates various networks has drawn out of the station, some passengers are eager in this trip, and others are pulled forward with reluctance and cry, scream, struggle and list all sorts of defects of IP; however, in spite of all the defects thereof, IP has been adopted as a industry standard, and no other technology, except for IP, has such a large potentiality and development space. (Abstracted from IP Convergence: Building the Future, by Susan Breidenbach, Network World, Aug. 10, 1998).
With the explosive increment of Internet services, the application range thereof has been extended to each field and each industry in the society. In the view point of telecommunication industry, more and more traditional telecommunication services employ IP for transmission, i.e., so-called Everything Over IP. The framework of the current telecommunication network will gradually turn from circuit switching and the networking technology thereof to a new framework based on packet switching, in particular, IP; and services over telecommunication network will turn from telephone service to data service.
TCP/IP Network Protocol
TCP/IP (Transmission Control Protocol/Internet Protocol) is a protocol most widely applied over the world at present, and the prevalence thereof is closely related to the impetuous development of Internet. Originally, TCP/IP is designed for the prototype of Internet, ARPANET, for providing a full set of protocols that are convenient and practical and can be applied on various networks. It is proved by facts that TCP/IP has accomplished its tasks, it makes network interconnection easy, and it enables more and more networks to participate in the network interconnection, thereby becoming a de facto standard of Internet.
Application Layer: application layer is a general term for all applications that users face. On this layer, there exist a lot of protocols from the TCP/IP protocol family to support different applications, and the implementation of many familiar Internet-based applications cannot be separated from these protocols. For example, HTTP protocol used in World Wide Web (WWW) access, FTP protocol used in file transmission, SMTP used in e-mail sending, DNS protocol used in domain name resolution, Telnet protocol used in remote logon and so on all belong to TCP/IP on the application layer; for users, patterned operating interfaces constructed by software are seen, but in fact, the above protocols are operated in the background.
Transmission Layer: the function of this layer is mainly to provide communication between applications, and on this layer, protocols from the TCP/IP protocol family include TCP and UDP.
Network Layer: network layer is a very crucial layer in the TCP/IP protocol family, which mainly defines the format of IP address, thereby data of different application types can be transmitted on the Internet smoothly, and IP protocol is a network layer protocol.
Network Interface Layer: this is the lowest layer of TCP/IP software, which is responsible for receiving an IP packet and sending it via a network, or receiving a physical frame from a network, extracting an IP datagram and delivering it to an IP layer.
How does IP implement network interconnection? Network systems and devices manufactured by various manufacturers, for example, Ethernet and packet switching network, etc., cannot intercommunicate with each other, the main reason is that the formats of the basic units (technically referred to as “frames”) of data transmitted by them are different. In fact, IP protocol is a set of protocol software consisted of software programs, and it unitedly converts various different “frames” into the format of “IP packet”, such conversion is a most important feature of Internet, i.e., a feature of “openness”, which makes all computers able to realize intercommunication on the Internet.
Then, what is “data packet”? And what feature does it have? Data packet is also a form of packet switching, that is, data to be transmitted are segmented into “packets” and then transmitted out. However, it belongs to “connectionless type”, that is, each “packet” is transmitted out as an “independent message”, so it is called “data packet”. Thus, before communication starts, no circuit needs to be connected first, and respective packets will not necessarily be transmitted via one and the same route, so it is called “connectionless type”. Such a feature is very important, and in the case of text information transmission, it greatly improves the robustness and security of the network.
Each data packet has two parts, header and message. Header contains necessary contents such as destination address, etc., so that each data packet can correctly reach its destination via different routes. At the destination, the data packets recombine and restore to the data sent originally. This requires that IP has the functions of packet packaging and assembling.
During the practical transmission process, a data packet also needs to change the data packet length according to the packet size specified by the network it passes, the maximum length of an IP data packet may reach 65535 bytes.
Quality of Service (QoS) is a main problem of IP Internet. Through the ages, countless research reports try to solve this problem; however, if we arrange the main milestones of QoS in time order, it will be readily seen that this is a helpless history in which QoS of Internet continuously lowers its requirements and continuously fails. From “Inte Serv” (1990) to “Diff Serv” (1997) and then to “Lightload” (2001), the summation of various partial QoS improving solutions that seem effective is still far from the target of network-wide QoS. QoS seems nearby, but in fact it's too far away to reach.
At the early stage of IP Internet, video application has become a target of network service, for example, MBone. Due to the lack of an effective QoS, no video communication service with a commercial value can be developed in a long term, which weakens the profit-earning capacity of IP Internet. Therefore, it has a great commercial value to solve the quality problem of network transmission. The quality problem of network transmission specifically appears as packet loss and error code. Computer files are not sensitive to errors in transmission; so long as there exists a TCP retransmission mechanism, a computer may consider the network as usable even if a great part of data packets are lost during the transmission process. However, if packet loss rate and error code rate are higher than 1/1,000, the quality of video and audio will be lowered for synchronous video. Empirical data tells us that high-quality video communication even requires that packet loss and error code should be lower than 1/100,000. Test data from the current network environment show that most packet loss occurs inside a router, and error codes generated during optical fiber transmission may almost be neglected.
Why can't “Inte Serv” succeed?
“Inte Serv” is established on the basis of reserved independent stream resources by employing Resource Reservation SetupProtocol (RSVP). In large-scale network environment, if a part of bandwidth resources can be reserved between two video terminals, it may be specially used by the video service; however, although this sounds good, it is impracticable in fact.
Firstly, this solution requires network-wide device reconstruction, which equals to reestablishing the network, and it is almost impossible in practical operation.
Next, even if network-wide reconstruction is implemented, for example, a bandwidth of 2 Mbps is kept for a 2 Mbps video service in each switch, can QoS problem be solved? The answer is No.
The so-called 2 Mbps bandwidth of RSVP is only considered macroscopically, if data in one second is sent in the first half second centralizedly, a problem will arise and periodic burst flow will be formed. Because the core concept of IP Internet is “Best Efforts”, at each network node, the switch always tries its best to forward data at the highest speed. After a video stream passes multiple levels of switches, it will be certain that flow distribution becomes non-uniform. When multiple non-uniform and asynchronous streams are combined, greater non-uniformity will be generated in a period of time; that is, periodic congestion of network flow is certain to appear. With the increasing of video user numbers, no upper limit can be given to periodic congestion, and when it exceeds the internal storage capacity of the switch, packet loss will be directly caused.
Why does “Diff Serv” fail?
After “Inte Serv” made its appearance for 7 years, a novel method “Diff Serv” starts to prevail. “Diff Sew” tries to provide a network service being superior to “Best Efforts”. Such a method does not require complex network-wide resource reservation, and it is easy to implement. It only needs to put a “priority” label on each data packet and the network switch processes video data with “priority” first. The basic theory thereof is just like that a bank issues a gold card to a VIP client and the queuing time of a high-end client may be effectively reduced. This method also sounds good, but in fact, it is impracticable, too.
There exists one easy fact that cannot be ignored: the flow of a single video service is much larger than that of a traditional non-video service (over a hundredfold).
When there are a few video users, video data packets will be seen almost everywhere on the network. If most of the data packets have a “gold card”, VIP is meaningless. Additionally, because IP interconnection networking principle is not compulsory, although QoS has drawn up a set of moral standards for users that maintain their personal integrity during chaotic times, it is unpractical to require all the users to carry the standards into effect.
Therefore, “Diff Sew” is only effective in a few enterprise private networks, and it is difficult to be effectively popularized in large-scale public networks.
Why can't “Light load” succeed?
Since IP Internet was popularized step by step, people have been unremittingly seeking after an effective prescription for network QoS. After more than 10 years' brain squeeze, network technicians work out two QoS solutions, but neither is ideal. Under the macro-environment in which people loose confidence in solving QoS, some anonymous people put forward a method, i.e., “Light load”. The basic design consideration thereof is so-called light-load network, and it is considered that so long as a sufficient bandwidth is provided and optical fiber enters users' houses, there should be no need to worry about network congestion.
Is the design consideration of light-load network feasible? The answer is also No.
The current network technicians seem to miss a basic theory: the root of network packet loss phenomenon is flow non-uniformity. Macroscopically, when the sending speed is high in one time period, it is certain to cause jam in another time period; no upper limit can be given to the peak flow of the network so long as the network flow is non-uniform, and any arbitrary large bandwidth may be occupied in a short time.
Actually, a reasonably good video program may be transmitted so long as there exists a bandwidth of 2 Mbps; if a bandwidth of 8 Mbps is provided, a video content of HDTV quality may be transmitted. However, if we randomly browse a text or a picture on an ordinary web site, the instantaneous flow will be tens of folds of that of HDTV, because most of the current web site servers use a Gigabit network interface. If the flows of a lot of similar web sites just collide, the burst flow generated in a certain short time will exceed the flow required by all network-wide users that use HDTV, and a network with any bandwidth can be occupied. As shown by statistical analysis, such a collision is frequent.
IP Internet tries to absorb the instantaneous flow by employing a memory, which causes the increase of transmission delay. The storage capacity is limited, but the burst flow has no upper limit; therefore, by employing the memory method, it can only improve the packet loss of the current device, and the burst flow absorbed at the current node will put much pressure on the next node. Video stream flow is ceaseless, and the storage mode of the switch intensifies the accumulation of the burst flow to a weak node, thus network packet loss is inevitable.
By employing light load plus “Diff Serv” technology, the current network constructors may deal with narrowband VoIP voice services. This is because voice does not occupy the main part of total flow on the network; once jam occurs, voice will take priority by sacrificing computer files. However, for high-bandwidth video communication, only temporary improvement can be obtained by partial expansion. If expansion is also carried out at other nodes, the non-uniformity of network flow will increase therewith, so that the effect of the originally expanded part will be lowered. If expansion is carried uniformly network wide, the transmission quality will return to that before expansion. In other words, overall expansion is ineffective.
At preset, device manufacturers recommend ultra-wideband access networks of tens or even hundreds of Megabit to each household. However, even if optical fiber enters each household, it is difficult to exhibit a video communication service with good QoS to the consumers. In spite of what complex QoS measures are taken, the transmission quality of IP Internet can only be “improved”, and no quality of network transmission can be “guaranteed”.